Prefer listening?
Apple Podcasts | Spotify
Why do so many investigating a topic end up reaching the same conclusion?
Driving through Utah recently, I was struck by the smoke from the wildfires. The scale of the fires was staggering. Later that day, I watched a Facebook video of the governor announcing a statewide fireworks ban.
What surprised me wasn’t the announcement. It was the comments. Many weren’t about fireworks at all. They were comments about AI data centers using water.
That made me wonder:
Why were so many people making the same connection?
At first, I thought perhaps the topic of data centers had simply become popular. Then I realized something more interesting.
The First Question Shapes Everything That Follows
Recommendation algorithms don’t decide what is true - they decide what you’re likely to watch next. Recommendation systems also introduce another possibility.
Beyond simply recommending content based on our interests, platforms may make choices about which posts receive greater visibility and which posts receive less visibility.
Former employees at social media companies have described they can “amplify” or “throttle” content. Whether or not that occurred in this case isn’t something I can determine for certainty.
It reminded me that there are at least two forces shaping what we see:
- what we choose to engage with,
- and what the platform chooses to show us.
Suppose someone watches a video titled: “AI Data Centers Are Draining Our Water Supply.” The algorithm learns something simple: “This person is interested in AI, data centers, and water.” From that point forward, many of the recommended videos are likely to stay within that same frame.
More stories about AI.
More stories about water.
More stories about data centers.
The problem isn’t that the information is false.
The problem is that it begins with the same assumption: that AI data centers consume an unusually large amount of water.
At that point, few if any recommendation systems might ask a different question, for example, “What information would help this person better understand the issue?” Instead, they consider the question, “What content is this person most likely to watch next?” These are different objectives
.
The Missing Questions
As I investigated, I found my questions changing. Instead of asking, “Do data centers use too much water?” I found myself asking:
- How was the water use calculated?
- Does the number represent just cooling or is it a lifecycle water usage?
- Compared to what?
- Would it be more useful if we measured water per unit of computation, instead of water per unit of electricity?
- How much water is associated with charging an electric vehicle using the same metrics?
- How would terrestrial data centers compare to orbital data centers using the same lifecycle accounting?
Those questions led somewhere different than the original headlines.
The same thing happened when I investigated AI music. The popular question seems to be, “Is AI stealing music?” The question I eventually found myself asking was, “How did the recordings actually enter the training system?” That single change completely altered the direction of the investigation.
The Real Problem
I don’t think the biggest challenge is misinformation or recommendation algorithms. The larger problem is that our investigation never escapes the frame established by the first piece of content we watch.
We naturally search for more information. And the algorithm naturally recommends more of what appears relevant. Before long, we’re learning more and more...
...about the original assumption. Not about the broader problem.
Ironically, someone trying to investigate a topic can become increasingly informed while remaining inside the same conceptual frame.
Thinking Beyond the Feed
Sometimes thinking clearly requires more than consuming additional information. It requires stepping outside the original question. Instead of asking, “Why are AI data centers using so much water?” try asking, “How is water use actually calculated?”
Instead of, “Is AI stealing music?” ask, “How did the recordings enter the training system?” Those are different questions. Different questions lead to different investigations.
The more I investigate complex issues, the more I suspect that the quality of our conclusions depends less on how much information we consume, than on whether we’re willing to question the first question itself.
Recommendation algorithms may reinforce our assumptions. Editorial decisions may reinforce them. Even conversations with friends and family may reinforce them.
The pattern begins much earlier.
It begins with the first question we ask. Once that question becomes accepted, we spend our time looking for better answers…
...instead of asking whether we started with the right question in the first place.